This application claims the priority of Korean Patent Application Nos. 2002-602, filed on Jan. 5, 2002, and 2002-51883, filed on Aug. 30, 2002, in the Korean Intellectual Property Office, the disclosure of which is incorporated herein in its entirety by reference.
1. Field of the Invention
The present invention relates to image coding, and more particularly, to an image coding and decoding method and apparatus considering human visual characteristics.
2. Description of the Related Art
In general, an image is compressed for storage/transmission purposes. FIG. 1 shows a flowchart illustrating a conventional image compression method, wherein, in order to generate a compressed bit stream, spatial/temporal prediction coding (100), transform coding (110), quantization (120), and entropy coding (130), are sequentially carried out. In this case, most losses are generated during the quantization operation 120. This lossy compression method includes a lossy compression method for a still image and a lossy compression method for a moving image. JPEG is a representative lossy compression method for a still image, while MPEG-1, 2, 4, and H.261 and H.263 are representative lossy compression methods for a moving image.
Meanwhile, discrete cosine transform (DCT) is carried out when an image is coded. In this case, since the amount of calculation is too large in order to perform DCT on the whole image, an image is divided into blocks of a predetermined size, i.e., 8×8, and is then coded. Also, when quantization is performed, the amount of information increases if the image is coded using a quantization parameter for each unit block, so as to make a quantization parameter for each unit block different. Thus, the same quantization parameter is used in the whole image. In MPEG-4 and H.263 systems, information is provided in order to adjust a quantization parameter by ±2 for each block of 16×16. Then, the information is used to achieve the accurate target bitrate.
When such a coder is used, the image is displayed with similar image quality on the entire scene. However, when the user looks at an image, the user considers the image quality of a region of interest (ROI) more important than the image quality of a background region. This is why there is a difference between the regions the user can see at one time. The user intends to look at the region of interest (ROI) more carefully and to overlook other detailed portions of the background region. In particular, this phenomenon remarkably appears in case of a moving image. Thus, when an image is coded with less bits, improvement of the image quality of the region of interest (ROI) is needed by allocating more bits to the region of interest (ROI) than to the background region, rather than uniformly allocating the bits to the whole image.
In the MPEG-4 and H.263 systems, a part of an image is divided into regions and coded. In the MPEG-4 system, a user can define the regions in units of pixels using shape coding beyond a core profile. The above method is mainly used in MPEG-4 because operations can be performed in units of each object constituting a scene. Each of the objects is coded using different bitstreams, and user interaction can be performed in MPEG-4 using the above structure. Using this method, the ROI and the background region are separated from each other for each object such that the image is coded with different image quality. However, this object separation process is very complicated. Even though the objects are simply separated from one another using a rough shape, information is additionally needed in showing the shape of each of the objects, and thus a compression efficiency is lowered.
Also, in the H.263 system, a part of an image can be divided into regions in units of groups of consecutive macroblocks (MBs) or in units of groups of macroblocks (MBs) in a certain rectangular shape using a slice structured mode at an annex K and the image can be coded. This method, used in the H.263 system, is robust to errors. An important portion in an environment using a multiple transmission channel is transmitted via a transmission channel in a better environment such that a transmission efficiency is improved and errors occurring in a region are prevented from spreading into another region. In this case, the ROI can be coded using a slice structure in a rectangular shape. However, in order to show the background region, a part of an image must be divided into several rectangles, and thus, the structure of the H.263 system becomes complicated.
In U.S. Pat. No. 5,764,803 entitled by “Motion-adaptive Modeling Scene Content for very low Bit Rate Model assisted Coding of Video Sequences”, a part of an image is divided into a region of interest (ROI) and a background region and then, the image is coded. However, there is a limitation in the range of a quantization parameter which can be varied in each region. Thus, due to a difference in the image quality between the region of interest (ROI) and the background region, a boundary between the region of interest (ROI) and the background region can be seen.
Also, U.S. Pat. No. 6,263,022 entitled by “System and Method for fine granular scalable (FGS) Video with selective Quality Enhancement” discloses a compression method used in a multiple transmission channel environment including a base layer and an enhancement layer. The method can adapt to an environment of a transmission channel, but it is difficult to perform inter prediction, and thus a coding efficiency decreases. Also, the image quality of the region of interest (ROI) is improved, but an overall coding efficiency decreases. Thus, the image quality of the background region is greatly lowered. That is, a difference in image quality between the region of interest and the background region increases, and to this end, the boundary between the region of interest and the background region remarkably appears.
Also, U.S. Pat. No. 6,256,423 entitled by “Intra-frame quantizer Selection for Video Compression” discloses a compression method in which a region of interest (ROI) and a background region, a transition region between the region of interest (ROI) and the background region are defined and a quantization parameter between regions is determined. In the method, because of the transition region, a phenomenon by which a boundary between the region of interest (ROI) and the background region appears can be slightly prevented. However, there is a limitation in the range of the quantization parameter of each region, and n transition regions are also needed when n region of interests (ROls) exist in a part of a region, and thus a coding method is complicated. In addition, in order to smoothen the boundary between regions, another transition region between the transition region and another region is additionally needed. As a result, it is difficult to determine a quantization parameter of each region. In order to solve this problem, a method of iteratively selecting a quantization parameter has been also used, but this method results in an increase in the amount of calculation.